Sentiment Analysis of Political Tweets: Towards an Accurate Classifier

摘要

We perform a series of 3-class sentiment classification experiments on a set of 2,624 tweets produced during the run-up to the Irish General Elections in February 2011. Even though tweets that have been labelled as sarcastic have been omitted from this set, it still represents a difficult test set and the highest accuracy we achieve is 61.6% using supervised learning and a feature set consisting of subjectivity-lexicon-based scores, Twitter specific features and the top 1,000 most discriminative words. This is superior to various naive unsupervised approaches which use subjectivity lexicons to compute an overall sentiment score for a pair.